Nvidia’s First Microprocessor: HyperTransport, Silicon-on-Insulator and When It Makes Sense

Nvidia’s Central Processing Unit Might Be Well in Development

by Anton Shilov
03/17/2009 | 10:27 AM

Rumours about Nvidia Corp.’s possible entering the market of central processing units (CPUs) have been circulating since the year 2000, well before unified shader architecture or computing shaders, but the company has never confirmed such plans. But now that the intentions have been reaffirmed, it looks like Nvidia has been working on a microprocessor, or, at least, related technologies for quite some time now.


Graphics processing units (GPUs) and central processing units used to have almost nothing in common a decade ago, but huge advances in the graphics chip design as well as the evolution of CPUs have greatly moved both towards each other: GPUs are now much more programmable than back in the days, whereas microprocessors now feature many cores, built-in memory controllers and will soon even gain graphics engines. So, the same way that consumer electronics is converging with personal computer, CPUs are converging with GPUs.

Nvidia completely understands that convergence and has been preparing to enter the market of microprocessors as well as other chips for many years.

Firstly, Nvidia joined HyperTransport Consortium at the time when the technology was called Lighting Data Transport and was primarily known as a universal bus for next-generation CPUs by Advanced Micro Devices. This allowed Nvidia to get a bus for potential microprocessors as well as other chips and also let the company to enter the market of chipsets, first with AMD- and then with Intel-compatible core-logic.

Secondly, Nvidia entered Silicon-on-Insulator process technology consortium. While the technology can be used to produce various components and chips, it is primarily used to make complex leading-edge microprocessors based on x86, ARM or Power architectures.

Thirdly, back in January ’09 Nvidia replaced its chief scientist David Kirk with William Dally, who is known for his expertise in parallel computing technologies as well as work on Cell/Larrabee-like microprocessors. Some even claim that Nvidia has been hiring microprocessor specialists from Stexar for several years now, hence, the appointment of Mr. Dally may be just a part of wider CPU-related hiring program.

Now that the company presumably has microprocessor design specialists, the rights to use HyperTransport bus and some degree of expertise in SOI process technology, it seems clearly that Nvidia is working on CPUs or complex system-on-chip devices with a microprocessing unit inside. The question is when the company plans to release something that will compete against AMD and Intel. The answer to the question is simple: once AMD and Intel completely conquer the market of platforms by integrating crucial system components into their microprocessors there will be no place for Nvidia. Once this happens, it will make sense for the latter to begin promoting its own platforms with own CPUs or SoCs. It is very likely to happen rather sooner than later...